Improve route randomization

doc/lightning-getroute.7

@@ -2,12 +2,12 @@
 .\"     Title: lightning-getroute
 .\"    Author: [see the "AUTHOR" section]
 .\" Generator: DocBook XSL Stylesheets v1.79.1 <http://docbook.sf.net/>
-.\"      Date: 01/16/2018
+.\"      Date: 02/19/2018
 .\"    Manual: \ \&
 .\"    Source: \ \&
 .\"  Language: English
 .\"
-.TH "LIGHTNING\-GETROUTE" "7" "01/16/2018" "\ \&" "\ \&"
+.TH "LIGHTNING\-GETROUTE" "7" "02/19/2018" "\ \&" "\ \&"
 .\" -----------------------------------------------------------------
 .\" * Define some portability stuff
 .\" -----------------------------------------------------------------
@@ -31,7 +31,7 @@
 lightning-getroute \- Protocol for routing a payment (low\-level)\&.
 .SH "SYNOPSIS"
 .sp
-\fBgetroute\fR \fIid\fR \fImsatoshi\fR \fIriskfactor\fR [\fIcltv\fR]
+\fBgetroute\fR \fIid\fR \fImsatoshi\fR \fIriskfactor\fR [\fIcltv\fR] [\fIfuzzpercent\fR] [\fIseed\fR]
 .SH "DESCRIPTION"
 .sp
 The \fBgetroute\fR RPC command attempts to find the best route for the payment of \fImsatoshi\fR to lightning node \fIid\fR, such that the payment will arrive at \fIid\fR with \fIcltv\fR\-blocks to spare (default 9)\&.
@@ -41,6 +41,10 @@ There are two considerations for how good a route is: how low the fees are, and
 For example, if you thought there was a 1% chance that a node would fail, and it would cost you 20% per annum if that happened, \fIriskfactor\fR would be 20\&.
 .sp
 If you didn\(cqt care about risk, \fIriskfactor\fR would be zero\&.
+.sp
+The \fIfuzzpercent\fR is a floating\-point number, between 0\&.0 to 100\&.0, unit of percent\&. The \fIfuzzpercent\fR is used to distort computed fees along each channel, to provide some randomization to the route generated\&. 0\&.0 means the exact fee of that channel is used, while 100\&.0 means the fee used might be from 0 to twice the actual fee\&. The default is 5\&.0, or up to 5% fee distortion\&.
+.sp
+The \fIseed\fR is a string whose bytes are used to seed the RNG for the route randomization\&. If not specified, a random string is used\&.
 .SH "RISKFACTOR EFFECT ON ROUTING"
 .sp
 The risk factor is treated as if it were an additional fee on the route, for the purposes of comparing routes\&.

doc/lightning-getroute.7.txt

@@ -9,7 +9,7 @@ lightning-getroute - Protocol for routing a payment (low-level).
 
 SYNOPSIS
 --------
-*getroute* 'id' 'msatoshi' 'riskfactor' ['cltv']
+*getroute* 'id' 'msatoshi' 'riskfactor' ['cltv'] ['fuzzpercent'] ['seed']
 
 DESCRIPTION
 -----------
@@ -29,6 +29,19 @@ fail, and it would cost you 20% per annum if that happened,
 
 If you didn't care about risk, 'riskfactor' would be zero.
 
+The 'fuzzpercent' is a floating-point number, between 0.0 to 100.0,
+unit of percent.
+The 'fuzzpercent' is used to distort computed fees along each channel,
+to provide some randomization to the route generated.
+0.0 means the exact fee of that channel is used,
+while 100.0 means the fee used might be from 0 to twice the actual fee.
+The default is 5.0, or up to 5% fee distortion.
+
+The 'seed' is a string whose bytes are used to seed the RNG for
+the route randomization.
+If not specified, a random string is used.
+
+
 RISKFACTOR EFFECT ON ROUTING
 ----------------------------
 The risk factor is treated as if it were an additional fee on the route,

doc/lightning-pay.7

@@ -2,12 +2,12 @@
 .\"     Title: lightning-pay
 .\"    Author: [see the "AUTHOR" section]
 .\" Generator: DocBook XSL Stylesheets v1.79.1 <http://docbook.sf.net/>
-.\"      Date: 02/17/2018
+.\"      Date: 02/19/2018
 .\"    Manual: \ \&
 .\"    Source: \ \&
 .\"  Language: English
 .\"
-.TH "LIGHTNING\-PAY" "7" "02/17/2018" "\ \&" "\ \&"
+.TH "LIGHTNING\-PAY" "7" "02/19/2018" "\ \&" "\ \&"
 .\" -----------------------------------------------------------------
 .\" * Define some portability stuff
 .\" -----------------------------------------------------------------
@@ -36,12 +36,14 @@ lightning-pay \- Protocol for sending a payment to a BOLT11 invoice
 .sp
 The \fBpay\fR RPC command attempts to find a route to the given destination, and send the funds it asks for\&. If the \fIbolt11\fR does not contain an amount, \fImsatoshi\fR is required, otherwise if it is specified it must be \fInull\fR\&. If \fIbolt11\fR contains a description hash (\fIh\fR field) \fIdescription\fR is required, otherwise it is unused\&. The \fIriskfactor\fR is described in detail in lightning\-getroute(7), and defaults to 1\&.0\&. The \fImaxfeepercent\fR limits the money paid in fees, and defaults to 0\&.5\&. The \(oqmaxfeepercent\(cq is a percentage of the amount that is to be paid\&.
 .sp
+The \fBpay\fR RPC command will randomize routes slightly, as long as the route achieves the targeted \fImaxfeepercent\fR\&.
+.sp
 The response will occur when the payment fails or succeeds\&. Once a payment has succeeded, calls to \fBpay\fR with the same \fIbolt11\fR will succeed immediately\&.
 .sp
 When using \fIlightning\-cli\fR, you may skip optional parameters by using \fInull\fR\&. Alternatively, use \fB\-k\fR option to provide parameters by name\&.
 .SH "RETURN VALUE"
 .sp
-On success, this returns the payment \fIpreimage\fR which hashes to the \fIpayment_hash\fR to prove that the payment was successful\&.
+On success, this returns the payment \fIpreimage\fR which hashes to the \fIpayment_hash\fR to prove that the payment was successful\&. It will also return, a \fIgetroute_tries\fR and a \fIsendpay_tries\fR statistics for the number of times it internally called \fBgetroute\fR and \fBsendpay\fR\&.
 .sp
 On error, if the error occurred from a node other than the final destination, the route table will be updated so that getroute(7) should return an alternate route (if any)\&. An error from the final destination implies the payment should not be retried\&.
 .sp

doc/lightning-pay.7.txt

@@ -23,6 +23,9 @@ The 'maxfeepercent' limits the money paid in fees, and defaults to 0.5.
 The `maxfeepercent' is a percentage of the amount that is to be
 paid.
 
+The *pay* RPC command will randomize routes slightly, as long as the
+route achieves the targeted 'maxfeepercent'.
+
 The response will occur when the payment fails or succeeds.  Once a
 payment has succeeded, calls to *pay* with the same 'bolt11' will
 succeed immediately.
@@ -36,6 +39,9 @@ RETURN VALUE
 
 On success, this returns the payment 'preimage' which hashes to the
 'payment_hash' to prove that the payment was successful.
+It will also return, a 'getroute_tries' and a 'sendpay_tries'
+statistics for the number of times it internally called *getroute*
+and *sendpay*.
 
 On error, if the error occurred from a node other than the final
 destination, the route table will be updated so that getroute(7)

gossipd/gossip.c

@@ -2,6 +2,7 @@
 #include <ccan/cast/cast.h>
 #include <ccan/container_of/container_of.h>
 #include <ccan/crypto/hkdf_sha256/hkdf_sha256.h>
+#include <ccan/crypto/siphash24/siphash24.h>
 #include <ccan/endian/endian.h>
 #include <ccan/fdpass/fdpass.h>
 #include <ccan/io/fdpass/fdpass.h>
@@ -1047,15 +1048,29 @@ static struct io_plan *getroute_req(struct io_conn *conn, struct daemon *daemon,
 	u16 riskfactor;
 	u8 *out;
 	struct route_hop *hops;
-
-	fromwire_gossip_getroute_request(msg, &source, &destination,
-					 &msatoshi, &riskfactor, &final_cltv);
+	double fuzz;
+	u8 *rawseed;
+	struct siphash_seed seed;
+	size_t seedbytes;
+
+	fromwire_gossip_getroute_request(tmpctx, msg,
+					 &source, &destination,
+					 &msatoshi, &riskfactor, &final_cltv,
+					 &fuzz, &rawseed);
 	status_trace("Trying to find a route from %s to %s for %d msatoshi",
 		     pubkey_to_hexstr(tmpctx, &source),
 		     pubkey_to_hexstr(tmpctx, &destination), msatoshi);
 
+	/* Initialize siphash */
+	memset(&seed, 0, sizeof(seed));
+	seedbytes =
+		(tal_len(rawseed) > sizeof(seed)) ?	sizeof(seed) :
+		/*otherwise*/				tal_len(rawseed) ;
+	memcpy(&seed, rawseed, seedbytes);
+
 	hops = get_route(tmpctx, daemon->rstate, &source, &destination,
-			 msatoshi, 1, final_cltv);
+			 msatoshi, 1, final_cltv,
+			 fuzz, &seed);
 
 	out = towire_gossip_getroute_reply(msg, hops);
 	tal_free(tmpctx);

gossipd/gossip_wire.csv

@@ -107,6 +107,9 @@ gossip_getroute_request,,destination,struct pubkey
 gossip_getroute_request,,msatoshi,u32
 gossip_getroute_request,,riskfactor,u16
 gossip_getroute_request,,final_cltv,u32
+gossip_getroute_request,,fuzz,double
+gossip_getroute_request,,seedlen,u16
+gossip_getroute_request,,seed,seedlen*u8
 
 gossip_getroute_reply,3106
 gossip_getroute_reply,,num_hops,u16

gossipd/routing.c

@@ -3,7 +3,6 @@
 #include <bitcoin/block.h>
 #include <bitcoin/script.h>
 #include <ccan/array_size/array_size.h>
-#include <ccan/crypto/siphash24/siphash24.h>
 #include <ccan/endian/endian.h>
 #include <ccan/structeq/structeq.h>
 #include <ccan/tal/str/str.h>
@@ -331,10 +330,24 @@ static u64 risk_fee(u64 amount, u32 delay, double riskfactor)
 
 /* We track totals, rather than costs.  That's because the fee depends
  * on the current amount passing through. */
-static void bfg_one_edge(struct node *node, size_t edgenum, double riskfactor)
+static void bfg_one_edge(struct node *node, size_t edgenum, double riskfactor,
+			 double fuzz, const struct siphash_seed *base_seed)
 {
 	struct node_connection *c = node->in[edgenum];
 	size_t h;
+	double fee_scale = 1.0;
+
+	if (fuzz != 0.0) {
+		u64 scid = short_channel_id_to_uint(&c->short_channel_id);
+		u64 h =	siphash24(base_seed, &scid, sizeof(scid));
+
+		/* Scale fees for this channel */
+		/* rand = (h / UINT64_MAX)  random number between 0.0 -> 1.0
+		 * 2*fuzz*rand              random number between 0.0 -> 2*fuzz
+		 * 2*fuzz*rand - fuzz       random number between -fuzz -> +fuzz
+		 */
+		fee_scale = 1.0 + (2.0 * fuzz * h / UINT64_MAX) - fuzz;
+	}
 
 	assert(c->dst == node);
 	for (h = 0; h < ROUTING_MAX_HOPS; h++) {
@@ -345,7 +358,7 @@ static void bfg_one_edge(struct node *node, size_t edgenum, double riskfactor)
 		if (node->bfg[h].total == INFINITE)
 			continue;
 
-		fee = connection_fee(c, node->bfg[h].total);
+		fee = connection_fee(c, node->bfg[h].total) * fee_scale;
 		risk = node->bfg[h].risk + risk_fee(node->bfg[h].total + fee,
 						    c->delay, riskfactor);
 
@@ -380,7 +393,9 @@ static bool nc_is_routable(const struct node_connection *nc, time_t now)
 static struct node_connection *
 find_route(const tal_t *ctx, struct routing_state *rstate,
 	   const struct pubkey *from, const struct pubkey *to, u64 msatoshi,
-	   double riskfactor, u64 *fee, struct node_connection ***route)
+	   double riskfactor,
+	   double fuzz, const struct siphash_seed *base_seed,
+	   u64 *fee, struct node_connection ***route)
 {
 	struct node *n, *src, *dst;
 	struct node_map_iter it;
@@ -440,7 +455,8 @@ find_route(const tal_t *ctx, struct routing_state *rstate,
 					SUPERVERBOSE("...unroutable");
 					continue;
 				}
-				bfg_one_edge(n, i, riskfactor);
+				bfg_one_edge(n, i, riskfactor,
+					     fuzz, base_seed);
 				SUPERVERBOSE("...done");
 			}
 		}
@@ -1118,7 +1134,8 @@ struct route_hop *get_route(tal_t *ctx, struct routing_state *rstate,
 			    const struct pubkey *source,
 			    const struct pubkey *destination,
 			    const u32 msatoshi, double riskfactor,
-			    u32 final_cltv)
+			    u32 final_cltv,
+			    double fuzz, const struct siphash_seed *base_seed)
 {
 	struct node_connection **route;
 	u64 total_amount;
@@ -1130,6 +1147,7 @@ struct route_hop *get_route(tal_t *ctx, struct routing_state *rstate,
 
 	first_conn = find_route(ctx, rstate, source, destination, msatoshi,
 				riskfactor / BLOCKS_PER_YEAR / 10000,
+				fuzz, base_seed,
 				&fee, &route);
 
 	if (!first_conn) {

gossipd/routing.h

@@ -2,6 +2,7 @@
 #define LIGHTNING_LIGHTNINGD_GOSSIP_ROUTING_H
 #include "config.h"
 #include <bitcoin/pubkey.h>
+#include <ccan/crypto/siphash24/siphash24.h>
 #include <ccan/htable/htable_type.h>
 #include <ccan/time/time.h>
 #include <gossipd/broadcast.h>
@@ -186,7 +187,9 @@ struct route_hop *get_route(tal_t *ctx, struct routing_state *rstate,
 			    const struct pubkey *source,
 			    const struct pubkey *destination,
 			    const u32 msatoshi, double riskfactor,
-			    u32 final_cltv);
+			    u32 final_cltv,
+			    double fuzz,
+			    const struct siphash_seed *base_seed);
 /* Disable channel(s) based on the given routing failure. */
 void routing_failure(struct routing_state *rstate,
 		     const struct pubkey *erring_node,

gossipd/test/run-bench-find_route.c

@@ -177,6 +177,7 @@ int main(int argc, char *argv[])
 	struct pubkey me = nodeid(0);
 	bool perfme = false;
 	const double riskfactor = 0.01 / BLOCKS_PER_YEAR / 10000;
+	struct siphash_seed base_seed;
 
 	secp256k1_ctx = secp256k1_context_create(SECP256K1_CONTEXT_VERIFY
 						 | SECP256K1_CONTEXT_SIGN);
@@ -194,6 +195,7 @@ int main(int argc, char *argv[])
 	if (argc > 3)
 		opt_usage_and_exit("[num_nodes [num_runs]]");
 
+	memset(&base_seed, 0, sizeof(base_seed));
 	for (size_t i = 0; i < num_nodes; i++)
 		populate_random_node(rstate, i);
 
@@ -212,6 +214,7 @@ int main(int argc, char *argv[])
 		nc = find_route(ctx, rstate, &from, &to,
 				pseudorand(100000),
 				riskfactor,
+				0.75, &base_seed,
 				&fee, &route);
 		num_success += (nc != NULL);
 		tal_free(route);

gossipd/test/run-find_route-specific.c

@@ -127,7 +127,7 @@ int main(void)
 	nc->flags = 1;
 	nc->last_timestamp = 1504064344;
 
-	nc = find_route(ctx, rstate, &a, &c, 100000, riskfactor, &fee, &route);
+	nc = find_route(ctx, rstate, &a, &c, 100000, riskfactor, 0.0, NULL, &fee, &route);
 	assert(nc);
 	assert(tal_count(route) == 1);
 	assert(pubkey_eq(&route[0]->src->id, &b));

gossipd/test/run-find_route.c

@@ -108,7 +108,7 @@ int main(void)
 	/* A<->B */
 	add_connection(rstate, &a, &b, 1, 1, 1);
 
-	nc = find_route(ctx, rstate, &a, &b, 1000, riskfactor, &fee, &route);
+	nc = find_route(ctx, rstate, &a, &b, 1000, riskfactor, 0.0, NULL, &fee, &route);
 	assert(nc);
 	assert(tal_count(route) == 0);
 	assert(fee == 0);
@@ -123,7 +123,7 @@ int main(void)
 	status_trace("C = %s", type_to_string(trc, struct pubkey, &c));
 	add_connection(rstate, &b, &c, 1, 1, 1);
 
-	nc = find_route(ctx, rstate, &a, &c, 1000, riskfactor, &fee, &route);
+	nc = find_route(ctx, rstate, &a, &c, 1000, riskfactor, 0.0, NULL, &fee, &route);
 	assert(nc);
 	assert(tal_count(route) == 1);
 	assert(fee == 1);
@@ -138,22 +138,22 @@ int main(void)
 	add_connection(rstate, &d, &c, 0, 2, 1);
 
 	/* Will go via D for small amounts. */
-	nc = find_route(ctx, rstate, &a, &c, 1000, riskfactor, &fee, &route);
+	nc = find_route(ctx, rstate, &a, &c, 1000, riskfactor, 0.0, NULL, &fee, &route);
 	assert(nc);
 	assert(tal_count(route) == 1);
 	assert(pubkey_eq(&route[0]->src->id, &d));
 	assert(fee == 0);
 
 	/* Will go via B for large amounts. */
-	nc = find_route(ctx, rstate, &a, &c, 3000000, riskfactor, &fee, &route);
+	nc = find_route(ctx, rstate, &a, &c, 3000000, riskfactor, 0.0, NULL, &fee, &route);
 	assert(nc);
 	assert(tal_count(route) == 1);
 	assert(pubkey_eq(&route[0]->src->id, &b));
 	assert(fee == 1 + 3);
 
 	/* Make B->C inactive, force it back via D */
 	get_connection(rstate, &b, &c)->active = false;
-	nc = find_route(ctx, rstate, &a, &c, 3000000, riskfactor, &fee, &route);
+	nc = find_route(ctx, rstate, &a, &c, 3000000, riskfactor, 0.0, NULL, &fee, &route);
 	assert(nc);
 	assert(tal_count(route) == 1);
 	assert(pubkey_eq(&route[0]->src->id, &d));